DNSMOS (ONNX)

Non-intrusive speech quality prediction from Microsoft's Deep Noise Suppression Challenge. Mirrored unmodified for use by speechonnxmetrics.

Upstream: microsoft/DNS-Challenge, also distributed in the speechmos PyPI package. Licence: MIT (see LICENSE.txt).

Files

file	input	output
`sig_bak_ovr.onnx`	raw waveform `[N, 144160]` @ 16 kHz	`[N, 3]` — SIG, BAK, OVRL (P.835)
`model_v8.onnx`	`[N, 900, 120]`	`[N, 1]` — P.808 MOS
`sig.onnx`	`[N, 901, 161]`	`[N, 1]`
`bak_ovr.onnx`	`[N, 901, 161]`	`[N, 3]`
`personalized/pdnsmos_sig_bak_ovr.onnx`	raw waveform `[N, 144160]` @ 16 kHz	`[N, 3]`

Raw model outputs require the upstream polynomial score mapping to become MOS values.

Reddy et al., DNSMOS: A Non-Intrusive Perceptual Objective Speech Quality Metric to Evaluate Noise Suppressors, ICASSP 2021.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support